Operator Guide

What is an Operator?

An operator is an independent infrastructure provider that deploys a Tangle Blueprint to serve AI model inference. Operators stake TNT as collateral, set their own pricing, and earn a share of the inference revenue they serve.

How Routing Works

When a user sends a request for a model, the Tangle Router finds all operators serving that model and routes to the best one based on latency, uptime, and price. If an operator fails, the request automatically retries with the next-best operator.

Pricing & Revenue

You set per-token input and output prices for each model you serve. The platform takes a fee and you keep the remainder. Payments settle on-chain — no invoicing, no waiting.

Reputation & Slashing

Your reputation score is based on uptime, latency, and request success rate. Higher reputation means more traffic routed to you. Sustained downtime or SLA violations can result in stake slashing.

Requirements

• GPU hardware (NVIDIA A100/H100) or Modal/cloud account
• TNT stake to register on-chain (slashable on SLA violations)
• Publicly accessible HTTPS endpoint
• Health check endpoint at /health
• Low-latency responses (routing favors faster operators)

Deploy & Register

1. Pick or build a Blueprint. Deploy a vLLM, Ollama, TGI, or custom inference target. Your endpoint must serve the OpenAI-compatible /v1/chat/completions route and expose a /health check.
2. Deploy on Tangle. Use the Tangle CLI to publish your Blueprint and stand up a service instance on testnet:

   cargo tangle blueprint deploy

3. Register & stake on-chain. CallregisterOperator() on the Blueprint Service Manager, stake TNT collateral, and set your endpoint URL and per-token pricing on-chain.
4. Get discovered. The gateway's chain-sync picks up your operator automatically. Once your first health check passes, the router starts ranking you against live latency, uptime, and price and routing traffic your way.